CN103646404A - Color-based rapid tea flower segmenting and counting method - Google Patents

Abstract

The invention relates to the technical field of tea leaves and image processing, and discloses a color-based fast tea tree flower segmentation and counting method. The present invention firstly converts the original RGB color image of tea leaves into HSI color space, and performs image enhancement processing, performs aggregation calculation based on the characteristic hue on the hue H therein, and then converts the image back into RGB color space; then applies the improved fast The region growing and merging algorithm performs preliminary seed selection according to the R and G parameters of tea tree flowers, and performs region growth on the seed region based on color similarity and region adjacency, and combines color distance and merging rules for region growth and merging , and finally complete the segmentation and counting of tea tree flowers. Experimental results show that the algorithm has good connectivity and can easily and quickly segment multiple tea tree flowers from tea images.

Description

A Color-Based Fast Tea Tree Flower Segmentation and Counting Method

technical field

The invention belongs to the technical field of tea and image processing, and relates to a method for segmenting and counting tea tree flowers in tea trees based on color enhancement combined with an improved region growing and merging algorithm.

Background technique

The development and application of tea tree flower is a hot direction in recent years. The ingredients in tea tree flower have the functions of detoxification, antibacterial, hypoglycemic, anti-aging and enhancing immunity. Its protein, tea polysaccharide, tea polyphenol, active antioxidant The substance exceeds the content of similar substances in tea, while the pesticide residue and heavy metal content are very low.

A tea tree can generally grow nearly a thousand vegetative buds for tea in a year, and at the same time, it can also grow more reproductive flower buds for breeding offspring. With the popularization and application of asexual reproduction technology, reproductive flower buds are no longer responsible for the task of breeding offspring, and a large number of tea tree flowers have become a burden to compete with tea tree buds and leaves for water and fertilizer. In order to ensure the quality and quantity of tea leaves in the coming year, tea farmers spend The method of artificial pruning and spraying plant growth regulators should be taken to suppress the reproduction and growth of tea tree flowers.

Now tea tree flowers or dried flowers are made into puree and compound powder through deep processing, which can be added to food, beverage, daily cosmetics, women's and children's hygiene products and other products. It has good functions and is a rare natural product. Composite raw materials. At the same time, harvesting flowers in winter slack can increase agricultural income and increase tea production in the coming year.

For color images, the main segmentation algorithms include edge detection methods, clustering methods, and region-based methods. For color images, edge detection methods often cannot accurately locate objects and objects in the image, and clustering methods are too computationally intensive. , the region growing method can directly act on the color space, and the image color distribution and regional connectivity etc. can be fully considered in the algorithm. Therefore, in the segmentation and counting of tea tree flowers of the present invention, the region growing method is selected. However, region growing algorithms are often affected by the selection of initial seed points and the order of growth, and there are also problems in determining the rules of region growing and merging.

Aiming at the color features of the tea tree flower color image in the tea image, the invention proposes a simple and fast color-based tea tree flower segmentation and counting method. This method first converts the original RGB tea image through the image color space, and enhances the characteristics of the tea tree flower in the H hue in the HSI color space, and in the subsequent segmentation, an improved region growing algorithm is used. On the basis of the growth algorithm, the selection method of the seed point, the growth rule and the merge rule are improved. Experimental results prove that this method is not only simple and efficient, but also consistent with human vision while ensuring regional connectivity.

Contents of the invention

Usually an initial color image is generated in the RGB color space, which is represented by three components of R, G, and B. The RGB color component has a good effect on color display, but it is generally not suitable for color analysis, because there is a great correlation between the three components of R, G, and B. But for the tea tree flower image, because the tea tree flower is white, it is very prominent in the tea tree image based on green and dark green, and the purpose of the present invention is to quickly detect the tea tree flower, so as to facilitate the detection of the tea tree flower at an appropriate time Picking, the experiment found that the two color components of R and G of tea tree flowers are obviously different from the true colors of tea leaves and tea buds. Therefore, in the present invention, firstly, the original tea RGB color image is converted into the color space, and the image is converted from the RGB color space to the HSI color space, and the characteristic tone H of the tea tree flower is image-converged to enhance processing to reduce The level of color is convenient for the later region growth and merging, and then the image color space is converted back to the RGB color space; then the improved fast region growth and merging algorithm is applied, and the initial seed selection is performed according to the R and G parameters of the camellia flower. The seed area is grown based on the similarity of color and the adjacency of the area, and the final growth and merging of the area is carried out in combination with the color distance and the merging rules, and finally the rapid segmentation and counting of the tea tree flowers are completed.

A kind of fast tea tree flower segmentation and counting method based on color of the present invention, the main process of its algorithm is:

(1) Obtain the original image of the tea leaves;

(2) Convert the original image from the RGB color space to the HSI color space, and perform image enhancement convergence calculation based on the characteristic hue for the hue H in it;

(3) Convert the image back to RGB color space;

(4) According to the R and G parameters of the tea tree flower, select some feature pixels in the image as seeds;

(5) Based on the growth rules, the seed area is grown, and adjacent pixels with similar color properties to the seed are attached to the seeds of the growth area;

(6) Based on the merging rules and color distance, scan and select multiple sub-block areas of the entire image, and merge areas that are similar in color and adjacent in space;

(7) Perform morphological processing of expansion and contraction on the merged region;

(8) Complete the segmentation and counting of multiple tea tree flowers.

The value range of H in the HSI color space is [0,360], and in the tea image, the tea tree flower has prominent white, and the stamen then has prominent yellow, the hue image enhancement method that the present invention proposes is: define the tea tree petal and The center values of the two characteristic hues of the stamen, on the plane of the hue circle, the experiment found that when the two hue values correspond to 30 and 43 respectively, better results can be achieved, so centering on these two values, the center that satisfies the condition The hue values near the value are converged to the central value according to different conditions and step lengths, so as to reduce the number of relatively secondary hues, make each hue value closer to the central hue, and achieve the purpose of making the original image have a stronger sense of hierarchy. It is also more conducive to the next step of the division of tea tree flowers.

The step size is very important for the selection of convergence. If the step size is too large, it will cause adverse effects such as sudden color changes in the image. If the step size is too small, the adjustment effect will not be obvious. After many experiments and tests, it is determined that the step size is selected as three different convergence step sizes of 1, 3, and 5. The step size close to the center value of the hue is small, and the step size farther away from the center value of the hue is larger. The pooled computation of hue is defined as:

H _i '=H _i ±k

1<|H _i -H ₀ |≤5 k＝±1

Among them, 5<|H _i -H ₀ |≤10 k＝±3

10<|H _i -H ₀ |≤15 k＝±5

H _i is the hue value of a certain point in the image, H _i ′ is the hue value after converging calculation, H ₀ is the hue center value, k is the step size, when H _i <H ₀ , the value of k is positive, when H _i When >H ₀ , the value of k is negative.

In the process of seed selection and region growth, the seed pixel is selected as the starting point of growth according to the eigenvalue condition in the image, and then the pixels with the same or similar properties as the seed pixel in the 8 neighborhoods around the seed pixel are merged into this region , and then use these new pixels as new seed pixels to continue the above process until no pixels satisfying the conditions can be included. At the same time, considering the connectivity and adjacency between different pixel points, the seed points under each cycle traversal are dynamically changed during the region growing process.

When performing region growth on the 8 neighborhoods around the seed point, the region growth rules are defined:

|rr ₀ |≤k _r ,|gg ₀ |≤k _g

Among them, r ₀ , g ₀ are the R and G components of any point P ₀ selected on the image, and r and g are the R and G component values of any point in the 8 neighborhoods around P ₀ , k _r , k _g is the threshold selected corresponding to the R and G components.

Experiments show that the growth rule defined in this way not only reduces the computational complexity and speeds up the segmentation time, but also fully considers the gradient between each color component, which is more conducive to the extraction of color features of color images.

In the area merging of images, the present invention defines that two areas are similar in color, adjacent in space, and there is no significant edge in their neighborhoods, which are two connectable areas, that is, the distance between an area and its adjacent area The maximum value of the relative color distance is smaller than the defined threshold. The present invention adopts the color component mean value of the region to define the color distance for calculation, and the color distance is defined as follows:

${\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>$

和

代表两个区域的颜色分量均值，||||表示欧式距离。r_i·r_j的乘积使得包含像素数目较少的区域与其他区域的颜色距离相比较小，从而在颜色分量均值相同的情况下，有利于小区域的优先合并，使得分割结果更加符合人们的视觉特性。Where r _i and r _j represent the number of pixels contained in the i and j regions respectively,

and

Represents the mean value of the color components of the two regions, and |||| represents the Euclidean distance. The product of r _i r _j makes the color distance between the region with a small number of pixels smaller than that of other regions, so that when the mean value of the color components is the same, it is conducive to the preferential merging of small regions, making the segmentation results more in line with people's expectations visual properties.

In the region growing process, if the color distance between the selected seed region and its neighbors is less than the preset distance threshold, the two regions can be merged, and each neighborhood is retraversed after the merger, and then judged whether the region is within the threshold Within the range, if it is less than the threshold, the regions will be merged intermittently until there is no more similar regions to merge.

And at the end of the region merge, require:

If the number of pixels in an area is less than a certain threshold and the color distance meets the relevant conditions, then this area is merged into its nearest neighbor area.

At the same time, in the process of selecting sub-block regions of tea tree flowers, there may be similar small regions on old leaves or buds, so if the number of pixels in a region is less than a certain threshold and there is no adjacent mergeable region , remove it.

At the end of the segmentation, the morphological algorithm processing method is used to expand and shrink the segmented image respectively to reduce small holes. Finally, the tea tree flowers separated after completion are counted and counted.

Description of drawings

Figure 1 is a flow chart of the fast tea tree flower segmentation and counting algorithm.

Figure 2 is the original image of the tea leaves.

Figure 3 is the segmentation result of tea tree flower

Detailed ways

Below in conjunction with accompanying drawing, the implementation item of the segmentation and counting method of fast tea tree flower is specifically described.

The digital camera used in the experiment is CANON S80. In the process of image acquisition, close-range mode is adopted, and the flashlight is turned off to avoid the influence of the light of the flashlight on the color of the tea image. At the same time, the image should be taken under natural light and avoid direct sunlight. The focal length of the imaging taken in the imaging is about 30cm, and the resolution is 1600×1200.

The experiment found that when the focal length of the image is closer, the number of tea tree flowers obtained is less, and the accuracy of segmentation is higher, and the accuracy of segmentation is the highest when there are 10 to 20 tea tree flowers. When the focal length is closer, it will affect the depth of field and imaging clarity, thereby affecting the accuracy of segmentation. But when the focal length is the farthest, the imaging image of tea tree flowers will be reduced, the range of image color will be wider, and the accuracy of segmentation will also be improved. decline.

k _r and k _g are the thresholds selected corresponding to the R and G components. When the thresholds and color distances of the R and G components are given, the experiment finds that k _r =9, k _g =7, and D _c takes the value When it is 10, better results are obtained.

The original image format of a digital camera is RGB format, but there is a strong correlation between the three components of R, G, and B. With the change of lighting conditions, the three components of R, G, and B will have great changes. Directly use these The component often cannot get the desired effect, so in the selection of the image color space, the HSI space is selected.

Under the HSI model, the color information of the image is mainly reflected by H and S, and the conversion formula from RGB to HSI space is as follows:

$\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}& \mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; G</span>}\\ \mathrm{<span\; class="notranslate">\; \&theta;</span>}& \mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; G</span>}\end{array}\right.$

in $\mathrm{\&theta;}=\arccos \left[\frac{[(R-G)+(R-B)]/2}{\sqrt{{(R-G)}^2+(R-B)(G-B)}}\right]$

$\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 3</span>}\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; B</span>}}$

$\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; )</span>$

The experimental results show that the segmentation of the algorithm has good connectivity. The experiment is based on a computer with a host CPU of Intel Core Dual-Core I3-3220 and a memory of 4G. Tea tree flower segmented from tea leaves image. It provides a good method for tea farmers to quickly obtain tea tree flower information. But at the same time, due to the rapid segmentation based on color, some tea tree buds cannot be segmented because they are turquoise.

Claims (3)

1. the quick Tea Flower based on color is cut apart and a method of counting, it is characterized in that comprising following concrete steps:

(1) obtain the original image of tealeaves;

(2) by original image from RGB color space conversion to HSI color space, and the figure image intensifying that tone H wherein carries out based on feature tone is converged to calculating;

(3) image is converted back to RGB color space;

(4), according to the R of Tea Flower, G parameter, in image, select Partial Feature pixel as seed;

(5) based on growth rule, seed region is grown, the neighbor similar to seed color character is attached on the seed of growth district;

(6) based on merging rule and color distance, a plurality of sub-blocks region of entire image is scanned and chosen, to close in color, region adjacent on space merges;

(7) expand and the morphology that shrinks is processed in the region after being combined;

(8) complete cutting apart and counting of a plurality of Tea Flowers.

2. a kind of quick Tea Flower based on color according to claim 1 is cut apart and method of counting, it is characterized in that: the figure image intensifying of tone H being carried out based on feature tone in step (2) is converged in calculating, two Tea Flower feature TINTCs have been defined, near the tone value central value satisfying condition is converged to central value by different conditions and step-length, its step-length elects 1,3 as, 5 three kind different converges step-length, step-length near TINTC is little, more larger away from the step-length of TINTC, calculating is defined as:

H _i′＝H _i±k

1＜|H _i-H ₀|≤5 k＝±1

Wherein, 5 < | H _i-H ₀|≤10 k=± 3

10＜|H _i-H ₀|≤15 k＝±5

H _ifor the tone value of certain point in image, H _i' converge the tone value after calculating, H ₀for TINTC, k is step-length, works as H _i< H ₀time, k value, for just, is worked as H _i> H ₀time, k value is for negative.

3. a kind of quick Tea Flower based on color according to claim 1 is cut apart and method of counting, it is characterized in that: in the growth of step (5) seed region, defined Rule of Region-growth:

|r-r ₀|≤k _r,|g-g ₀|≤k _g

Wherein, r ₀, g ₀selected any point P on image ₀r, G component, r, g are P ₀point is R, the G component value of the interior any point of 8 neighborhoods around, k _r, k _gfor corresponding to R, the selected threshold value of G component.

Images